Stock feature dimensionality reduction for closing price prediction using unsupervised machine learning technique (case study of Nigeria Stock Exchange)

Abstract

Stock data offers invaluable insights into the world of finance. It encourages investment and savings for both individuals and the nation as a whole. To monitor and predict stock, many stock variables are collected which in turn leads to curse of dimensionality because they occupy much storage space and take more computational time. In order to avoid this, there is a need to reduce the dimensionality of the stock features. Since stock is an unlabeled data with a Gaussian distribution (the features are normally jointly distributed), an unsupervised machine learning technique was applied to discover, establish an association and extract the most important features that have the entire generality of the original dataset for predicting the next day’s closing price. The dataset (daily price list) of Dangote Sugar Refinery Plc was randomly selected from the 27 blue chip companies in Nigeria Stock Exchange. 4 stock features were discovered and extracted from the 9 features in the original dataset.